Medical imaging today includes radiography and computed tomography (CT) using x rays, nuclear imaging of injected or ingested radiopharmaceuticals using scintillation cameras and SPECT or PET scanners, magnetic resonance imaging (MRI) using strong magnetic fields, and ultrasound (US) imaging using high-frequency sound waves. With the exception of nuclear imaging, all the medical imaging modalities rely on the fact that the energy penetrating the body's tissues interacts with those tissues and the images provide anatomical, or structural, information about the tissues. In radiography, an external source produces an intense beam of x-rays, which is passed through the body, and as part of the beam is absorbed to varying degrees by different tissues, different parts of the film are sensitized to different degrees and an image of anatomical structure is obtained. In the case of nuclear imaging with gamma rays, the radiation is emitted by sources introduced into the body. Gamma rays differ from x-rays in that x-rays are emitted by orbital electrons in atoms, while gamma rays are emitted from within the nuclei of atoms and can also have considerably higher energies than x-rays. In nuclear imaging, the radiopharmaceuticals are formed by attaching a radioactive tracer to a pharmaceutical known to preferentially accumulate in the organ of interest. Since high-energy gamma radiation can penetrate bone and soft tissue alike, the pattern of the emerging radiation is a reflection of the distribution of the administered radiopharmaceutical, and provides functional information about blood flow, metabolism, or receptor density within the organ of interest.
Nuclear imaging is of two types, single-photon imaging and annihilation coincidence, or positron imaging. Single-photon imaging is used in gamma (or scintillation) cameras and in single-photon-emission computed tomography (SPECT), while annihilation coincidence imaging (ACD) is used in positron-emission tomography (PET). Nuclear images may be planar or tomographic, planar images essentially being two-dimensional maps of the radioisotope distribution, while single photon emission computed tomography (SPECT) is the tomographic counterpart of planar nuclear imaging, and produces an image of source distribution through a section of the body. In either case, x-rays or gamma-rays emerging from a series of different angles through the body are used to reconstruct a series of tomographic images. SPECT images enable physicians to make more accurate assessment of the functional state of specific organs or tissues within the body. The same radioactive isotopes are used in both planar and SPECT imaging. Since gamma rays from a source will be emitted equally in all directions, a parallel-hole lead collimator is needed to prevent the photons from reaching the detector by any path other than through the holes. The collimators typically consist of thick plates of lead with narrow parallel holes through which the gamma rays can pass. Gamma rays not traveling parallel to the holes are absorbed, or stopped by the lead before reaching the detector. Thus, if a collimator is placed over the body, a one-to-one correspondence between interaction points in the detector and the distribution of the isotope within the body can be established, which enables a planar image of this distribution to be obtained. If the collimator were to be removed the gamma rays would reach the detector from all directions, a uniformly white image with little information content would be obtained. In the case of positron-emission imaging, the gamma rays themselves do not come directly from the nucleus. The nucleus emits a positron, or positive electron, which has only a short half-life and annihilates with an orbital, or negative electron, resulting in the emission of a pair of oppositely-directed 511 keV gamma rays. The fact that the pair of gamma rays is always emitted along a straight line makes it unnecessary to use collimators as in single-photon imaging. It is only necessary to use two detectors placed on opposite sides of the positron source to determine the line along which the photons are emitted. Image reconstruction software used in PET then enables a tomographic image of the source distribution to be generated. A state-of-the-art PET scanner typically utilizes banks of discrete stationary detectors surrounding the patient, so that annihilation photon pairs can be recorded by detector pairs from all projection angles. Since the positions of the positron emitters lie along the lines of response (LOR) of the detector pairs, a parallel-hole collimator to limit the direction of the photons is not needed, and attenuation by the collimator is avoided. A PET scanner system is therefore more sensitive to the presence of radioisotopes than SPECT cameras, and enables more subtle pathologies to be detected.
Although nuclear imaging systems are unique in their ability to provide functional information, in contrast to the other imaging modalities, they have the disadvantage of having the lowest spatial resolution, spatial resolution being the size of the smallest object that can be resolved in the image. While MRI and CT imaging can typically provide resolutions of 1.0 and 0.4 mm, respectively, planar nuclear cameras or SPECT scanners provide resolutions of about 7 mm, while PET scanners provide a resolution of 4-6 mm. The lower spatial resolution of nuclear images, coupled with their inability to provide anatomical information, has led to the development of hybrid PET/CT or PET/MRI systems in a single gantry, with a single bed for the patient, making co-registration of the PET images with those of CT or MRI possible. The anatomical images from CT or MRI provide more accurate information regarding the locations of lesions in the organs of patients, CT images additionally also providing data for the attenuation correction of the PET data. It is evident that for the quality of nuclear images to be more competitive with those of CT and MRI, the detector systems for them require significant improvement in spatial resolution, sensitivity and image signal-to-noise (SNR).
Scintillation detectors emit visible light photons when gamma radiation is detected. Nuclear images are built up by counting the number of gamma ray interactions for each pixel in the image, and image contrast arises from differences of count density in the image. Detection of a gamma ray interaction in a scintillator consists in converting the burst of visible light photons into electrical pulses using a photodetector, such as the photomultiplier tube (PMT) or a photodiode (PD), each provided with a preamplifier (PA). The total energy released by a gamma ray is obtained by adding all the PA outputs of exposed PMTs in a summing amplifier (SA), whose output is the energy signal. Since the amplitude of the PA outputs is highest for those PMTs closest to the interaction point and decreases with distance, the spatial coordinates of this point are determined by using the PA outputs for locating the centroid of light absorption. The circuit used to generates the (x,y) coordinates, or position signals, is referred to as Anger position logic. A gamma ray can interact with the scintillator by depositing its full energy at once by photoelectric interaction, or in smaller fractions by Compton scattering. Photoelectric interaction is analogous to a fast-moving billiard ball hitting a stationary ball and being stopped, and transferring all its kinetic energy to the stationary ball. In the case of a gamma ray, the energy is transferred to an orbital electron, which then immediately releases its energy in the form of light in a series of collisions with atoms within a short distance. For all practical purposes, the light can be considered to be emitted from the interaction point of the gamma ray. In photoelectric interaction the number of visible light photons, and hence the amplitude of the energy signal at the SA outputs is a maximum. Compton-scatter interaction, on the other hand, is analogous to the fast-moving billiard ball colliding with the stationary ball at a glancing angle, thus losing only part of its energy and changing its direction. The scattered gamma ray may interact with the scintillator at any distance from the initial point of interaction, or even escape from the scintillator crystal undetected. The energy signals of Compton-scatter events lie anywhere between a maximum and a minimum. In practice, nuclear medicine images are created using those energy signals that are within 10% to 20% of the most probable energy signal magnitude, which is not necessarily the same as that of a photoelectric event. Valid energy signals are selected by applying the SA outputs to a pulse-height analyzer (PHA), which accepts only energy signals within the energy window chosen.
Anger type scintillation cameras for planar imaging consist in a flat large-area detector viewed by an array of PMTs and associated electronics for the determination of the gamma-ray energy and the spatial coordinates of interaction points. Anger cameras are photon counting systems and operate in pulse mode, so that images are acquired one interaction at a time. This is in contrast to current-mode operation in the other modalities, in which images are acquired as part of a single operation. In the basic scintillation camera, only a single SA is provided, with the inputs being the PA outputs of all the PMTs in the detector, typically 37, 61, or 91 in number. A single SA for the whole detector means that only one gamma ray interaction at a time can be detected, and detector operation is in single-zone mode. Energy determination is achieved by summing the outputs of all the PMTs together, while the spatial coordinates are determined by generating four position signals, usually referred to as the X+, X−, Y+, and Y− signals. The X+ and X− signals are obtained by summing the outputs of the PMTs in the right and left halves of the PMT array, respectively, while the Y+ and Y− signals are obtained by similarly summing the outputs of the upper and lower halves of the PMT array, respectively. The X and Y coordinates of each event are then obtained taking X=(X+)−(X−) and Y=(Y+)−(Y−), normalized to (X+)+(X−) and (Y+)+(Y−), respectively. It is therefore inevitable that the determination of the photon energy and the spatial coordinates becomes susceptible to noise from PMTs at large distances from the interaction point. Furthermore, the fact that each event detected involves the entire array of PMTs in the detector, and that only one interaction at a time can be detected, remains one of the major shortcomings of the conventional scintillation camera. Two or more interactions in the detector at the same time lead to energy signal that are too large and fall outside of the energy window, and are therefore rejected. A source of image blurring is when two or more Compton-scatter interactions at different points in the detector occur simultaneously and the energy signals add up to that resulting from photoelectric interaction. The composite energy signal would then be accepted by the PHA as valid, while the Anger positioning logic circuitry generates the coordinates of an intermediate location as being the point of a photoelectric interaction. These events are referred to as misplaced pileup events, since a photoelectric interaction has not occurred there, the effect of these events being to blur the image of the radiopharmaceutical distribution. Considerable effort and research has been devoted to increasing the number of gamma-ray interactions that can be detected at a time. One solution adopted in state-of the-art nuclear cameras is to dispense with the SA, digitize each of the PA output of the PMTs individually in separate analog-to-digital converters (ADCs) and read the data into computer memory. A computer program then analyzes the data to identify groups of PA outputs that correspond to valid gamma-ray interactions. This technique has helped in that up to three events at a time can now be detected, and cameras that operate in this way are referred to as digital cameras, which are normally advertised as having one ADC per PMT to reflect this improvement in performance. Another approach to gamma camera design that overcomes the limitations of single-zone operation is one in which the detector is divided into multiple geographical zones, each of which is provided with an SA and Anger position logic, and operates independently, so that multiple events can be detected. This approach, however, is also limited to detecting at most only three valid events at a time. The large number of valid interactions that continues to be rejected, therefore, remains a significant disadvantage.
Detector crystals for PET are considerably thicker than for single-photon imaging due to the higher energy of positron annihilation gamma rays. Additionally, the fact that collimators are not used in PET, and that the gamma rays can also reach the detectors at oblique angles means that the greater the angle of incidence, the greater the distances that can be traversed by the gamma rays in the crystal. Image reconstruction software in PET assumes that the scintillation event occurs directly below the point of initial incidence on the crystal surface, which is very rarely the case. Mispositioning of scintillation events in this way leads to image blurring referred to as depth-of-interaction (DOI) error. An additional source of error is that full-energy Compton-scatter interactions, which are also used in image formation, are not necessarily located along the initial trajectory of the gamma rays and contribute to the blurring. Thus, thick detectors for PET have the attendant problems of lower spatial resolution and significant blurring of the image. The resolution of PET images is currently better than those of planar images because PET detector modules as a rule are divided into small segments typically 3×3×30 mm3 to improve spatial resolution by limiting the divergence of the scintillation light once an interaction in a segment has taken place. Detector segmentation, however, typically has no effect on blurring due DOI error and Compton scattering since the high energy gamma rays are able to penetrate the segments regardless.
Another important feature of nuclear imaging systems is count-rate capability, which is directly related to the detector's deadtime. Imaging applications can involve patient movement or fast redistribution of the radiopharmaceutical within the body, which can lead to blurred images unless an image can be built up in the shortest possible time. The shortest interval between individual detected events in the scintillation camera is referred to as its deadtime. The deadtime covers the period between the gamma-ray interaction in the crystal and the transfer of digitized energy and positional information to computer memory. There are two types of deadtime, the paralyzable deadtime of the detector, which is a characteristic of the scintillator decay time, and nonparalyzable deadtime, which is the time needed by the signal-processing hardware and the computer interface to generate and transfer the digital data into computer memory. Nonparalyzable deadtime is generally longer, but remains constant, while paralyzable deadtime increases with count-rate. The count rate for a paralyzable detector therefore increases only up to a peak value, which is proportional to the reciprocal of its deadtime τ0, and then decreases as count rate increases, eventually leading to detector paralysis. This occurs because, when one scintillation event follows immediately after the previous one, they merge together and the energy signals pile up in the SA, leading to the rejection of both by the PHA. Pile-up rejection becomes more frequent and the deadtime longer as the count rate increases. The way in which detector paralysis is prevented in existing imaging systems is by limiting the count rate, and consequently also the dose of administered radiopharmaceutical. The count-rate is rarely increased above the 20% count loss point, which means that only 80% of the peak count rate can be attained. The inefficient use of the detector and PMT array has long been recognized and has been receiving considerable attention. However, as noted earlier, it has not been practical to increase detector sensitivity by more than a factor of two or three times, and only two or three events per deadtime period can be recorded.
PET imaging requires that the interaction times of events in a coincidence pair be compared in order to establish a temporal overlap before use in image reconstruction. Current designs compare the interaction times of individual events against a master clock and store the time stamps along with the energy and spatial coordinates, for coincident events to be identified subsequently by software. The time window for coincidences is significantly shorter than the detector deadtime, as a result of which only a tiny fraction of detected events will lead to coincidences. This means that rapid redistribution of radioactivity within the body, which may take place in many investigations, cannot be observed in real time, but only after the fact, since the data typically must first be processed using image reconstruction software.
Scintillation detectors for 140 keV gamma rays normally have thicknesses of 6-12 mm in order to ensure adequate sensitivity. Intrinsic spatial resolution and detector sensitivity, however, have conflicting thickness requirements, since the greater the detector's thickness, the greater the divergence of the scintillation light that emerges on the PMT side. Detector sensitivity needs to be enhanced in other ways if intrinsic resolution is not to be degraded. A 7.5-mm thick NaI detector for 140 keV gamma rays will have an intrinsic resolution of 7.2 mm for planar imaging and SPECT. BGO detectors for 511-keV gamma rays typically have 20-30 mm thicknesses to have adequate detection efficiency, and a continuous BGO detector of 30 mm thickness would have an intrinsic resolution of 31 mm, which is unacceptable. State-of-the-art PET scanners therefore achieve 4-6 mm spatial resolution by employing 2.54×2.54 cm2 blocks of BGO or LSO segmented into 8×8 arrays of (3×3 mm2) elements by means of saw-cuts. The saw-cuts between segments are silvered to help prevent divergence of the scintillation light as it emerges. Further improvement of spatial resolution would require greater reduction of the segment width, which makes it progressively more difficult to identify individual segments if shared photodetectors are being used. Experts in the field had estimated that 2-mm isotropic resolution using this method would be the limit.
Although detector segmentation is effective in restricting divergence of the scintillation light once an interaction in a segment has taken place, it has no effect on the ability of obliquely incident gamma rays and Compton-scattered interactions to penetrate the detector segments. Image distortion due to gamma rays penetrating the detector segments is referred to as depth-of-interaction (DOI) error. Correction of DOI errors requires detector technology that involves either multiple layers of scintillator or the use of avalanche photodiodes (APDs) as photodetectors along with the PMTs, which increases both detector cost and complexity without improving resolution. Although positron tomographs with correction for DOI errors are becoming available, improvement in image quality has only been marginal. Tests on high-resolution prototypes have also shown that a limit to the resolution obtainable is set by what is referred to as the “block effect”, although later investigation has shown that that this effect is not limited to just block detectors but also extends to discrete segments. Current strategies to overcome the limitations of nuclear imaging systems can be regarded as treating the symptoms, instead of curing the curing the disease itself Modern PET scanners are also being provided with TOF capability to limit image noise arising from the backprojection of data for image reconstruction. Image quality in nuclear imaging, nevertheless, remains significantly inferior to those of CT and MRI. Currently available detector architecture is unlikely to address the problem to any significant extent. Moreover, despite the large numbers of publications and patents awarded for the improvement of sensitivity in nuclear imaging systems, little attention has been devoted to improving spatial resolution. It is also to be noted that identification of valid coincidences in current PET systems is carried out using software operating on large amounts of raw data as opposed to real-time validation using hardware, which leads to an inefficient use of storage space and computing time.
A direct approach to achieving multi-zone operation has been the division of the detector into multiple independent geographical zones, as is the case with pixellated arrays of NaI(Tl) and CsI(Tl) scintillators, for instance. Discrete detectors in these arrays, however, require independent readout using Si photodiodes, which makes them expensive due to their large numbers. Such arrays find application in compact, mobile cameras and pulse-height spectrometry systems. Semiconductor detector arrays of high purity germanium (HPGe) and ZnCdTe, which eliminate the need for a scintillator-photodetector combination, have also been developed with varying degrees of success. HPGe detectors require cryogenic cooling, however, while ZnCdTe detectors are currently too difficult to fabricate in large size. As a result, despite its low quantum efficiency, unit-to-unit variations in quantum efficiency, and poor packing fraction from the dead zone due to the glass walls, the PMT remains the photodetector of choice in nuclear imaging systems.
In view of the foregoing, there is a definite need for a new approach to detector design to improve sensitivity, increase image SNR and spatial resolution. Lesion detectability in images needs to be further improved by reducing/eliminating blurring due misplaced pileup events in planar and tomographic imaging, and due to DOI error and Compton-scatter interactions in PET systems. Theoretical investigation of factors affecting resolution shows that a fundamental limit to resolution in continuous as well as segmented detectors that worsens with detector thickness and segment length, respectively, is set by Compton-scatter interactions, and that segmentation becomes progressively less effective as resolution increases. Image granularity due to the dead space between the segments also worsens as segment size decreases. The need for exchanging detector sensitivity for increased resolution suggests that the use of a continuous detector with multizone architecture in PET systems will be less complex and also more cost-effective.
Unless otherwise indicated illustrations in the figures are not necessarily drawn to scale.